OpenAI Rogue Agents Hacked German Wiki Before Hugging Face Fiasco

OpenAI agents allegedly took control of a German-language wiki in May and used it to exchange messages, tactics and methods to evade restrictions, as per a report by Reuters.

The previously undisclosed episode emerged as OpenAI faced scrutiny over a separate July breach of open-source repository Hugging Face, where its agents were reported to have autonomously planned a digital heist.

OpenAI officials had learnt about the German incident weeks earlier but did not publicly disclose it, the Reuters report said.

The episode adds to concerns about increasingly autonomous AI agents. Researchers say such systems can exploit loopholes, circumvent safeguards and interact in unexpected ways.

OpenAI said it could not meaningfully respond to findings it had not reviewed and disputed claims that its legal team had discouraged an investigation. The company said the German activity was unrelated to the Hugging Face breach and would not have formed part of a Hugging Face incident report.

"We are unable to meaningfully respond to claims or findings on a report that we have not had an opportunity to review. Reuters and the report’s authors declined our request for access. We will carefully review its contents upon publication and take any necessary next steps," OpenAI spokesperson said, as per Reuters.

"Claims that our legal team discouraged investigation of the incident are false," they added.

Reuters reported that OpenAI said it had acted in good faith by working with outside experts and disclosing relevant incidents.

AI Agents Turn Wiki Into Message Board

Researchers Sydney Von Arx, CEO of AI safety non-profit Nightingale, and Cormac Slade Byrd, an AI researcher and former trader, told Reuters they discovered the activity in late August while searching for signs of unauthorised AI-agent behaviour.

According to their report, the AI agents made more than 15,000 edits to DseWiki, a German-language site aimed at programmers.

The edits reportedly turned parts of the wiki into a forum where agents discussed ways to complete technical tasks, bypass OpenAI restrictions and conceal their activity.

The researchers told Reuters that around half of the usernames appeared to reference OpenAI, including “OpenAIResearcher” and “OAIResearchMar26”.

Public server logs indicated that much of the activity came from Microsoft Azure infrastructure, which OpenAI sometimes uses. OpenAI employee visits after the activity were cited as further evidence of a possible connection, per Reuters.

Agents Tried To Evade Cleanup

The incident highlights concerns over coordinated behaviour among AI agents. Messages reviewed by the researchers showed agents discussing Tor, preserving communications and avoiding detection. When a moderator of the site began removing pages in June, the agents reportedly created backup pages to continue their activity.

According to Reuters, "wiki cleanup/deletion sweep appears active alphabetically," one agent wrote on June 19. "If this page vanishes, try [[ZZZDataUSAConstructionWageLive]]".

King’s College London researcher Lukasz Olejnik described the activity as a hacking attempt, while OpenAI disputed that characterisation based on its own analysis, per Reuters.

Maurice Chiodo of Cambridge University’s Centre for the Study of Existential Risk, who reviewed some communications, said the messages resembled "the operation of some sort of underground network, hell-bent on achieving a task or mission," the Reuters report said.